With the rapid development of artificial intelligence technology, the market structure of AI chips, as the core infrastructure supporting AI applications, is undergoing profound changes. Meta's acquisition of South Korean AI chip start-up FuriosaAI has undoubtedly become a key step in this change, indicating that Meta has officially challenged giants in the field of AI chips, such as Nvidia and Intel.
Meta's AI chip ambitions and strategic layout
As the world's leading social media and technology company, Meta has been actively exploring the application and innovation of AI technology. The acquisition of FuriosaAI is an important strategic layout of Meta in the field of AI chips. According to Forbes, Meta is in talks to acquire FuriosaAI, with a deal likely to be reached as early as February 2025. Meta has previously launched two generations of MTIA series AI inference chips, and if it successfully acquires FuriosaAI, it will further enrich its technical reserves in the field of AI semiconductors and enhance its competitiveness in the AI chip market.
Meta's strategic layout aims to gain a place in the AI chip market by acquiring FuriosaAI's advanced technology and talents, enhancing its hardware capabilities in the AI field, and reducing its dependence on external suppliers. This move will not only contribute to Meta's long-term development in the AI field, but will also have a profound impact on the competitive landscape of the global AI chip market.
Figure: Meta's acquisition of FuriosaAI to challenge Nvidia's AI chip hegemony
The technical advantages and market potential of FuriosaAI
FuriosaAI is a Seoul, South Korea-based AI chip start-up founded by Song Jae-joon, an executive and former CEO of gaming company Com2us, and currently led by June Paik, who previously worked at Samsung Electronics and AMD. FuriosaAI unveiled its second-generation AI accelerator RNGD at the 2024 Hot Chip 24, which uses a tensor shrink processor (TCP) architecture with excellent computing power and energy efficiency.
Figure: FuriosaAI's second-generation AI accelerator RNGD, which uses the TCP architecture and is said to have excellent computing power and energy efficiency
The RNGD is based on TSMC's 5nm process SoC, with an area of 653mm², a total of 40 billion transistors, 256MB of on-chip SRAM cache, and two 24GB HBM3 12Hi memory integrated in a CoWoS-S package, with an off-chip memory bandwidth of 1.5TB/s. RNGD AIC can provide 512 TOPS of INT8 computing power, 256 TOPS of BF16 computing power, or 1 POPS of INT4 computing power, while TDP power consumption is only 150W, supporting the deployment of high-end generative AI models such as Meta's Llama 3. FuriosaAI plans to achieve mass production of RNGDs in the second half of 2025.
FuriosaAI's technological strength lies in its innovative TCP architecture, which combines high performance, energy efficiency, and programmability, making it uniquely competitive in the AI chip market. RNGD is particularly outstanding, with significantly higher energy efficiency than NVIDIA's H100 and L40S when running large language models such as Llama 3.1 and GPT-J. For example, for a model with about 10 billion parameters, a single RNGD PCIe card can provide throughput performance of 2000~3000 tokens per second, depending on the context length. When running OpenAI's GPT-J 6B model, it can process about 12~15 queries per second.
Meta's strategic intent and market impact of the acquisition of FuriosaAI
Meta's strategic intent to acquire FuriosaAI is clear: to enhance its competitiveness in the AI field by acquiring advanced AI chip technology, reduce dependence on external suppliers, reduce hardware costs, and increase bargaining power. FuriosaAI's RNGD chip meets Meta's technical needs in the field of AI chips, especially in terms of supporting high-end generative AI models such as Llama 3.
The acquisition will give Meta a stronger position in the AI chip market, competing directly with giants such as Nvidia and Intel. Nvidia dominates the deep learning and data center AI computing market with its GPUs and Tensor cores, while Intel is highly competitive in AI inference with its Xeon processors and Movidius chips. The addition of Meta will bring new variables to the AI chip market and promote the reshaping of the market competition pattern.
The competitive landscape and development trend of the global AI chip market
The global AI chip market is highly competitive, mainly dominated by technology giants such as Nvidia, Intel, Google, and Amazon. According to the data of the China Research Institute of Puhua Industry Research Institute, the market size of the global AI chip industry has reached US$56.4 billion in 2023 and is expected to reach US$90.2 billion in 2024, with a compound growth rate of 24.55% in the next five years. This growth is mainly due to the rapid development of cloud computing, big data, Internet of Things and other fields, where the demand for AI chips is increasing.
With the continuous development of AI technology, the demand for AI chips is also growing. As an important infrastructure to support AI applications, the improvement of the performance and energy efficiency ratio of AI chips are of great significance to promote the popularization and application of AI technology. In the future, the AI chip market will usher in greater development opportunities, and technological innovation and market competition will become the main driving force for market development.
Challenges and opportunities of Meta acquires FuriosaAI
Meta's acquisition of FuriosaAI presents many challenges, but it also presents huge opportunities. The challenge is that the AI chip market is extremely competitive, and giants such as Nvidia and Intel have significant advantages in terms of technology, market, and ecosystem. Meta needs to make a comprehensive layout in terms of technology, market, and ecosystem to stand out from the competition.
The opportunity is that FuriosaAI's advanced technology will provide Meta with strong hardware support, making it more competitive in the field of AI chips. In addition, Meta can further expand its business in the AI field through the acquisition of FuriosaAI and promote the innovation and application of AI technology.
Conclusions
Meta's acquisition of FuriosaAI is an important strategic layout in the field of AI chips, which will provide strong technical support and market competitiveness for its long-term development in the field of AI. FuriosaAI's advanced technology will provide a strong guarantee for Meta to occupy a place in the AI chip market, and will also have a profound impact on the competitive landscape of the global AI chip market.
In the future, with the continuous development of AI technology, the AI chip market will usher in greater development opportunities. The addition of Meta will bring new vitality and innovation to the AI chip market and promote the reshaping of the market competition pattern. We look forward to Meta making greater breakthroughs in the field of AI chips and making greater contributions to the development of global AI technology."
